基于模糊神经网络的强化学习及其在机器人导航中的应用

Reinforcement learning based on FNN and its application in robot navigation

研究基于行为的移动机器人控制方法.将模糊神经网络与强化学习理论相结合,构成模糊强化系统.它既可获取模糊规则的结论部分和模糊隶属度函数参数,也可解决连续状态空间和动作空间的强化学习问题.将残差算法用于神经网络的学习,保证了函数逼近的快速性和收敛性.将该系统的学习结果作为反应式自主机器人的行为控制器。

Behavior-based robot navigation is studied. The fuzzy neural network（FNN）and reinforcement learning （RL） are integrated. RL is utilized for structure identification and parameters tuning of FNN. The problem of continuous, infinite states and actions in RL is solved by using the function approximationof FNN. Furthermore, the residual algorithm is applied to the FNN learning, which guarantees the convergence and rapidity. Then, the learning results are employed to design the controller of the reactive robot system, by which the problem of navigation under complicated environment is solved effectively.